The Artstar project is an autonomous creative AI. It proposes a technically grounded yet philosophically ambitious framework for building a self-evolving, multimodal creative agent—one capable of generating original visual, sculptural, and sonic outputs without reliance on external data sources.
At its core, the Artstar system operates as a closed generative loop composed of several integrated subsystems, each responsible for a specific modality (2D image, 3D form, and Music), with coordination orchestrated through a memory-augmented agentic logic layer. Rather than consuming or remixing datasets scraped from the internet, the agent begins from pure procedural randomness—initial seed prompts generated by a foundational LLM (e.g., GPT5) based on abstract conceptual primitives—and evolves through recursive self-reflection. The outputs of each generative cycle are re-ingested, described, and mutated, forming the basis for future iterations in a self-contained creative ecosystem. As far as the outputs are concerned there will be an initial period of prototyping and refining the output pipeline. There will eventually be a public facing version of Artstar as well. This could provide the resources for further development and could be structured in a variety of ways.
From a research perspective, the Artstar system aligns with open questions in AGI, especially around agent autonomy, self-reflective computation, and meta-cognitive planning. From an artistic standpoint, it represents a rupture from derivative machine art toward truly original, machine-native aesthetics. It is both a tool and an organism—designed to evolve, surprise, and challenge our assumptions about what it means for a machine to make art not from us, but from within itself.
The current Artstar project team is Steve Lomprey (Multimedia Artist and Founder) , Carl Bass (Former Head of Autodesk) , James Smith (Computer Scientist at UC Berkeley) , and advisory Josh Bloom (Professor of Astronomy at UC Berkeley, and AI startup founder).